problem with 3d registration

[esther1262]

Hi, I try to register two 3d labelmaps. The fixed one is a binary mask, dim 256x256x16. The moving one has 27 regions (values 0,1,...27), dim 256x512x256. Ultimately, I am only interested in a few regions.

img_mv = ants.image_read('results/templateLabels_27regions_256slices.nii.gz', dimension=3)
img_fx = ants.image_read('1.nii.gz', dimension=3) 

I think the pixel data are imported as float. Even if I specifed pixeltype='unsigned int' the registration results didnt improve.

mytx = ants.registration(fixed=img_fx , moving=img_mv, type_to_transform='BOLDAffine')
warped_moving = mytx['warpedmovout'] 

I tried all the possible choices for type_to_transform. None can improve the results.

img_fx.plot(axis=2, slices=11, overlay=warped_moving, title='After Registration (slice 11)', overlay_alpha=0.7)

This is one of the slices that has the worse result. Majority are okay, in terms of shape. Since I have a binary map, it's difficult to register all the regions. But I want the shape to be matched at least. This one has completely different shape.

Also,
moving labelmap after registration, so much noise

Anyone knows how to fix it? Thanks

[cookpa]

The correct arg is type_of_transform

Registering a binary mask to a label image is an unusual use case. There's very little information in a binary mask except at the edges. I would rethink this and reformulate the registration question if possible.

[esther1262]

The correct arg is type_of_transform

Registering a binary mask to a label image is an unusual use case. There's very little information in a binary mask except at the edges. I would rethink this and reformulate the registration question if possible.

typo, but they didnt make a difference

i have a mouse head mri, I want to register the labelmap template to the brain area. Thats why I created a binary map for the brain only, registration between a head and a brain image is even worse.

[cookpa]

typo, but they didnt make a difference

If there's a typo in the function call, the argument is ignored. The default will be used.

i have a mouse head mri, I want to register the labelmap template to the brain area. Thats why I created a binary map for the brain only, registration between a head and a brain image is even worse.

I agree you shouldnt' register head to brain. If you have a head + brain mask, you can create a brain image and register brain to brain, unless I'm missing something?

The noisy appearance in the output is from linear interpolation of the labels. You can use ants.apply_transforms with interpolator= genericLabel to avoid this.

[esther1262]

@cookpa we prefer not to change the pixel in brain images. we want to do it from the masks.